Semiconductor Watch: China's Chip Market Hits 90% Domestic
Last week we flagged China's chip market splitting into a separate supply chain, not just losing share to controls. This week, TrendForce put a number on how far that split has already gone: domestic suppliers now command an estimated 90% of China's high-end AI chip market. That's faster than most 2026 forecasts assumed — and it's reshaping how Nvidia's own $500 billion financing plan is being read by the analysts underwriting it.
01 — The Semiconductor Lead
TrendForce's August 10 supply-chain assessment finds China's high-end AI chip market shifting almost entirely to domestic solutions — a combination of homegrown GPUs and ASICs — leaving Nvidia, AMD, and other overseas suppliers with roughly 10% of sales this year. Separately, Bernstein's own forecast, cited by Fast Company and the Associated Press, puts Nvidia's China AI-chip share collapsing to about 8% by year-end (from roughly 40% in 2025) while Huawei rises to approximately 50%. The two estimates don't match exactly — they measure slightly different slices of the market — but they agree on direction and magnitude.
Limited H200 export approvals have pushed ByteDance, Tencent, Alibaba, and Baidu to accelerate domestic infrastructure investment, and analysts increasingly describe cluster stability, cost, and scalability — not single-card benchmark performance — as what actually determines whether Chinese silicon succeeds at this point. Nvidia's own SEC filing discloses $4.5 billion in charges tied to export restrictions in a single fiscal quarter this year, confirming the shift is now a material line item, not a rounding error. Not every analyst agrees this is settled: the Council on Foreign Relations has argued Huawei's aggregate compute output remains a small fraction of Nvidia's even at higher unit volumes, since China's strategy of producing larger quantities of less capable chips doesn't close the raw performance gap. Both things can be true — Huawei can be winning the domestic market on cost and availability while still trailing badly on frontier capability.
02 — AI Chip Watch
CoreWeave told Bloomberg this week it would face genuine difficulty if forced to shift meaningfully away from Nvidia chips — a rare, unusually candid admission from a major neocloud about how deep the CUDA software moat actually runs. It's a useful reality check against the custom-silicon-erosion narrative we've tracked in recent weeks: Broadcom and Marvell's ASIC gains remain real, but concentrated specifically in high-volume inference, not in the flexible training workloads CoreWeave's business depends on.
The scale of what's being built regardless of vendor is striking: Omdia now estimates global semiconductor industry revenue could jump 94% in 2026, with computing and data-storage chips alone approaching $1 trillion, while AMD separately projects the total high-performance and AI computing chip market could reach $2 trillion by 2030. Those are industry-wide totals, not company-specific claims — a reminder that even a company losing share in one geography can still grow in absolute terms as the overall market expands this fast.
03 — Memory & Packaging
Samsung's HBM4 yield is now confirmed to be holding near 80%, continuing last week's recovery trend as the company narrows the gap with SK Hynix. The more immediately visible effect of the memory squeeze is showing up somewhere most enterprise coverage misses: consumer graphics cards. Reports out of South Korea point to Nvidia GeForce RTX 50-series prices climbing as much as 30%, driven directly by higher TSMC wafer costs and GDDR7 memory pricing — with the RTX 5090 32GB already seeing significant increases in several markets, and similar pressure appearing in China and the U.S. Nvidia bundles GPU and memory as a package to card manufacturers, so GDDR7 cost inflation passes straight through to shelf price. The AI infrastructure memory shortage we've tracked for weeks is no longer just a hyperscaler procurement story — it's now a line item on ordinary consumer receipts.
04 — Foundry & Manufacturing
TSMC's board approved a $29.44 billion capital budget this week, alongside a legally binding joint venture with Sony Semiconductor Solutions to build next-generation image sensors. The new entity, Advanced Vision Semiconductor Manufacturing Corp, will be based in Kumamoto — the same region hosting TSMC's existing JASM logic plant — with Sony contributing roughly 465 billion yen ($2.92 billion) as controlling shareholder and TSMC adding 282 billion yen ($1.77 billion), targeting volume production in 2029. One analyst called it "virtually risk-free" for both sides: Sony caps its capital burden while securing TSMC's manufacturing expertise, and TSMC locks in steady foundry revenue from a customer rather than building sensor capability independently. Sony's CEO described the deal as "the first step towards a fab-light model" for the company.
On the core logic business, TSMC's 3nm output is projected to grow 20% by year-end versus the first half of 2026, even as demand for its 2nm node already runs meaningfully higher than 3nm — unsurprising given the performance and power efficiency gains, but a clear signal of where allocation pressure shifts next as 2nm ramps through Arizona and Kumamoto's second-phase capacity.
05 — Supply Chain & Geopolitics
The concentration risk cuts both ways this week. TSMC's dominance is itself becoming a structural vulnerability analysts are flagging more openly: as AI demand concentrates almost entirely on TSMC's advanced nodes and CoWoS packaging, competing foundries lose pricing power, and the industry's manufacturing know-how consolidates further into a single company — increasing the systemic cost if anything ever disrupts that one supply point. That's a different kind of concentration risk than the China bifurcation above, but it's compounding in the same direction: fewer, larger points of failure across the whole AI hardware stack.
06 — Capital & Competition
Nvidia's $500 billion Wall Street financing platform, covered here as a landmark move to securitize GPU compute, is now facing direct scrutiny over China exposure specifically — multiple analysts this week explicitly framed the plan's credit risk through the lens of Nvidia's collapsing China position, a link that wasn't part of the story when the platform was first announced. Nvidia is reportedly working to ease those credit-risk concerns directly with the consortium of lenders. Separately, AMD disclosed two high-severity TPM vulnerabilities affecting its memory controllers — a reminder that hardware-level security research is intensifying in parallel with the capital story, since flaws at the memory-controller level can undermine confidential computing guarantees enterprises are increasingly paying a premium for.
Semiconductor Market at a Glance
| Metric | Value |
|---|---|
| China High-End AI Chip Market, Domestic Share | ~90% |
| Nvidia China AI Chip Share, 2026 Forecast (Bernstein) | ~8% |
| TSMC New Capital Budget | $29.44B |
| TSMC-Sony Image Sensor JV | $4.69B |
| RTX 50-Series Price Increase (South Korea) | Up to 30% |
| Global Semiconductor Revenue Growth, 2026 (Omdia) | +94% |
07 — Three Semiconductor Signals
- China's chip market has functionally bifurcated already — TrendForce's 90% figure confirms the split happened faster than most 2026 forecasts modeled, not that it's still forming.
- Enterprise memory scarcity has reached consumer receipts — GDDR7 and wafer cost inflation are now visible in ordinary GPU pricing, not just hyperscaler procurement contracts.
- TSMC is hedging its own concentration risk through diversification — the Sony image-sensor JV extends TSMC into an adjacent AI-relevant market even as its core logic and packaging dominance goes structurally unchallenged.
THE CODEW TAKE
The question worth asking is no longer whether the global chip supply chain will split into a China-facing and a rest-of-world-facing track — this week's data says that's already substantially happened. The question now is how fast the two tracks diverge in capability, and TrendForce's cost-and-scalability framing versus CFR's raw-compute-gap argument represent genuinely different answers to that question, both grounded in real data.
For Nvidia specifically, that bifurcation is now inseparable from its own financing story — analysts explicitly tying credit risk on its $500 billion Wall Street platform to its collapsing China position shows the geopolitical and capital-markets threads that used to run in separate sections of this newsletter are now one story.
For technology buyers, the practical takeaway is that a single global AI hardware sourcing strategy no longer describes reality — Huawei's Ascend clusters and Nvidia's GPU platforms are increasingly incommensurable, and organizations with China operations need a genuinely separate hardware roadmap, not a regional variant of the same one.
Source Attribution
- TrendForce — China High-End AI Chip Market Supply Chain Observation, August 10, 2026
- MarketScale — Nvidia's China AI Chip Share Is Forecast to Collapse From 40% to 8% as Huawei Scales
- KuCoin News — China's AI Chip Market Nears 90% Domestic Share as Nvidia's Influence Declines
- Council on Foreign Relations — China's AI Chip Deficit: Why Huawei Can't Catch Nvidia
- Nvidia Corp — Form ARS FY2026 (SEC filing, export-control charges disclosure)
- Bloomberg — CoreWeave Warns of Difficulty if It Must Shift From Nvidia Chips
- Distill Intelligence — Semiconductors & AI Chips Weekly Briefing, August 14, 2026
- Reuters / Devdiscourse — TSMC Greenlights $29.44 Billion Expansion Amid AI Surge
- Sony Semiconductor Solutions — Sony and TSMC Agreed to Establish Joint Venture for Next-Generation Image Sensors
- LightReading — Sony and TSMC Establish $4.69B Image Sensor JV in Japan
- Taipei Times — TSMC Approves New Joint Venture With Sony
- FTC Electronics — Electronics Weekly News, Aug 3–9, 2026
- The Motley Fool — Demand for Chips Shows No Signs of Slowing (Omdia 94% forecast)
- Guru3D — AMD Publishes Fixes for Two High-Severity TPM Vulnerabilities
Editorial Note
The CODEW Semiconductor Watch tracks where structural leverage is shifting across the global chip industry — AI accelerators, memory, advanced packaging, foundry capacity, and the supply-chain and geopolitical forces reshaping who controls each bottleneck. It's built for technology leaders, investors, and supply-chain executives who need to see structural shifts before they become obvious market consensus.
Coverage is based on company disclosures, analyst research, industry reporting, and other publicly available information. Analyst estimates (e.g., Bernstein, TrendForce, Omdia) are attributed as forecasts, not confirmed figures, and may diverge from one another. Analysis reflects the reporting period and should be considered in the context of the sources and developments cited.
Reviewed by Erwin Castro
on
Tuesday, August 18, 2026
Rating: